CAZyme3D: a database of 3D structures for carbohydrate-active enzymes

Abstract

CAZymes (Carbohydrate Active EnZymes) degrade, synthesize, and modify all complex carbohydrates on Earth. CAZymes are extremely important to research in human health, nutrition, gut microbiome, bioenergy, plant disease, and global carbon recycling. Current CAZyme annotation tools are all based on sequence similarity. A more powerful approach is to detect protein structural similarity between query proteins and known CAZymes indicative of distant homology. Here, we developed CAZyme3D (https://pro.unl.edu/CAZyme3D/) to fill the research gap that no dedicated 3D structure databases are currently available for CAZymes. CAZyme3D contains a total of 870,740 AlphaFold predicted 3D structures (named Whole dataset). A subset of CAZymes 3D structures from 188,574 nonredundant sequences (named ID50 dataset) were subject to structural similarity-based clustering analyses. Such clustering allowed us to organize all CAZyme structures using a hierarchical classification, which includes existing levels defined by the CAZy database (class, clan, family, subfamily) and newly defined levels (subclasses, structural cluster [SC] groups, and SCs). The inter-family structural clustering successfully grouped CAZy families and clans with the same structural folds in the same subclasses. The intra-family structural clustering classified structurally similar CAZymes into SCs, which were further classified into SC groups. SCs and SC groups differed from sequence similarity-based CAZy subfamilies. With CAZyme structures as the search database, we created job submission pages, where users can submit query protein sequences or PDB structures for a structural similarity search. CAZyme3D will be a useful new tool to assist the discovery of novel CAZymes by providing a comprehensive database of CAZyme 3D structures.